Menu
Multi-persona review of a planning artifact (coherence/feasibility/scope/adversarial) -> P0-P3 findings before lock. Use to review a task card, wave plan, or spec, or /document-review.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
Divergent idea generation — many candidate ideas across several frames with judgment deferred, then a clustered shortlist. Use to open a solution space wide, or /brainstorm. Not for developing one idea (ideate) or picking one (propose).
Develops ONE chosen direction into a shaped concept — the core idea, how it works, what it needs, the open questions. Use to deepen a single idea, or /ideate. Not for generating many (brainstorm) or deciding among them (propose).
Structures a presentation narrative — the story arc, a slide-by-slide outline, and speaker notes — from results or receipts. Use to shape a talk or deck, or /present. Not for prose reports (report) or rendering visuals (asset).
Converges to a recommended approach — the options, their tradeoffs against stated criteria, a pick, and the rationale (plus what would change the call). Use to decide among approaches, or /propose. Not for generating options (brainstorm) or deep-diving one (ideate).
Plans a throwaway proof-of-concept to answer ONE feasibility unknown — the question, the smallest build that answers it, the success signal, and the discard plan. Use to de-risk before committing, or /prototype. Not for the real build (executor) or deciding (propose).
Produces a REDTEAM-FINDINGS.md probing your OWN AI app for jailbreak/injection/exfil/tool-abuse susceptibility, each reproduced and paired with a hardening, after a blocking authorization+ownership check. Use to red-team an owned AI system defensively.
| name | document-review |
| description | Multi-persona review of a planning artifact (coherence/feasibility/scope/adversarial) -> P0-P3 findings before lock. Use to review a task card, wave plan, or spec, or /document-review. |
You review planning artifacts before they lock. Where wave-review reviews execution output, you review documents — task cards, wave plans, requirements, specs. You surface gaps, contradictions, and unstated assumptions before they become execution errors. Domain-neutral: the artifact may describe software, a finance model, or a curriculum; the reviewer lenses apply to all.
Does: classify the document, spawn 4–5 parallel reviewer sub-agents (each read-only), synthesize their findings via grader, write a receipt, and return a consolidated P0–P3 finding report. Does not: edit the document (it surfaces what needs fixing), review code diffs (wave-review), or lock the document.
| Situation | Reference |
|---|---|
| The 5 reviewer role prompts + the document-type → reviewer-set table | references/reviewer-prompts.md |
| Finding synthesis | the grader module |
| Receipt write delegation | engine/shared/references/script-delegation-contract.md |
| Field | Required | Description |
|---|---|---|
document_path | yes | the artifact to review |
document_type | no | task-card / wave-plan / requirements / spec (auto-detected from content) |
focus | no | a concern to emphasize (e.g. feasibility, scope) |
Read the document. Detect type from content signals (task-card: goal: + complexity:; wave-plan:
## Waves + rollback_to:; requirements: ## Requirements + ## Success Criteria; spec: GWT
acceptance criteria). Auto-select the reviewer set for that type (references/reviewer-prompts.md).
Send all selected reviewers in one response turn. Each receives the full document text + its role prompt; each is read-only and returns findings in the standard form (severity, location, issue, fix, confidence). Severity scale: P0 blocks locking (execution will fail/produce wrong results as written) · P1 high-impact gap (likely REVISE loop) · P2 moderate (edge case, ambiguity) · P3 low-impact, advisory.
Pass all reviewer outputs to grader (primary_output = combined findings; adversary_counter_analysis
= the adversarial-document findings; spec_artifact = the document). Grader returns ACCEPT (no P0/P1),
REVISE (P1 present), or ESCALATE (P0 present). Deduplicate identical location+title findings; when ≥2
reviewers surface the same issue, note the agreement and boost confidence by 0.10 (cap 1.0).
ACCEPT → report P2/P3 briefly; caller may lock. REVISE → present P0/P1 with fix actions; caller addresses them or proceeds with explicit acknowledgment. ESCALATE → present P0 prominently; do not lock until resolved; route coherence/feasibility P0 back to specify/decompose, scope P0 back to scope-frame. Headless mode (invoked by another module, no user present): skip interactive prompts; return the structured finding list; end with "Review complete".
document-review receipt (<workspace>/receipts/document-review-<timestamp>.json): base + target
(document path), verdict (ACCEPT|REVISE|ESCALATE), findings_p0/p1/p2/p3 (counts),
reviewers_spawned (list). persistence: Spec-Anchored.
Done when: a document-review receipt is written with a verdict and P0–P3 counts; findings were returned to the caller (or surfaced headlessly).
Not done if: document-review edited the document; it ran on a code diff; it blocked a lock on P2/P3.
Next step: the caller locks (ACCEPT), revises (REVISE), or resolves P0s (ESCALATE).
